generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

GpuMat::setTo optimization (call cudaMemcpy2D if possible)

Browse Source
This commit is contained in:
Vladislav Vinogradov 2011-05-18 06:51:05 +00:00
parent b0598fcf36
commit f3b45af876
1 changed files with 16 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
modules/gpu/src/matrix_operations.cpp
View File

@ -442,6 +442,22 @@ GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask)
if (mask.empty()) if (mask.empty())
{ {
if (s[0] == 0.0 && s[1] == 0.0 && s[2] == 0.0 && s[3] == 0.0)
{
cudaSafeCall( cudaMemset2D(data, step, 0, cols * elemSize(), rows) );
return *this;
}
if (depth() == CV_8U)
{
int cn = channels();
if (cn == 1 || (cn == 2 && s[0] == s[1]) || (cn == 3 && s[0] == s[1] && s[0] == s[2]) || (cn == 4 && s[0] == s[1] && s[0] == s[2] && s[0] == s[3]))
{
int val = saturate_cast<uchar>(s[0]);
cudaSafeCall( cudaMemset2D(data, step, val, cols, rows) );
return *this;
}
}
typedef void (*set_caller_t)(GpuMat& src, const Scalar& s); typedef void (*set_caller_t)(GpuMat& src, const Scalar& s);
static const set_caller_t set_callers[8][4] = static const set_caller_t set_callers[8][4] =
{ {